Fast harmonic estimation using real-time embedded

Erhan Sezgin*, Anurag Mohapatra†, Vedran Peric†

†Center for Combined Smart Energy Systems (CoSES),

Munich School of Engineering

*Gazi University, Ankara

Fast harmonic estimation using real-time embedded controllers in CoSES smart grid

• Motivation

• Harmonic analysis methods

• Real time processing on HIL targets

• Conclusions and future work

2

Outline


MSE Colloquium | 30. July 2020

What are harmonics:Integer & non-integer(inter-harmonics) multiples of

fundamental component(50Hz) of a voltage or current

signal.

Caused by:• Power electronic devices

• Non-linear loads

• Unbalanced grids due to high penetration of

renewables.

Important for:• Grid monitoring

• Application of legal sanctions defined by regulations

• Active power filtering

4

Motivation

Fundamental, harmonics & overall signal.



How to provide accurate feedback of harmonic presence in the measurements for

microgrid experiments?

4

Motivation – within CoSES




microgrid experiments…while not burdening the real-time controller?

5






3




226 current & voltage measurements,

sampled at maximum 10kHz



3






6 embedded controllers

to accquire the signals



3






6 embedded controllers

to accquire the signals

Processor has many other

tasks in real-time

5

Harmonic analysis methods



50Hz, Fundamental

Mag. = 1, Phase = 0°

150Hz, 3rd Harm.

Mag. = 0.5, Phase = 270°

500Hz, 10th Harm.

Mag. = 0.1, Phase = 0°

350Hz, 7th Harm.

Mag. = 0.4, Phase = 30°

5




50Hz, Fundamental


150Hz, 3rd Harm.

Mag. = 0.5, Phase = 270°

500Hz, 10th Harm.

Mag. = 0.1, Phase = 0°

350Hz, 7th Harm.

Mag. = 0.4, Phase = 30°

5




50Hz, Fundamental


150Hz, 3rd Harm.

Mag. = 0.5, Phase = 270°

500Hz, 10th Harm.

Mag. = 0.1, Phase = 0°

350Hz, 7th Harm.

Mag. = 0.4, Phase = 30°

5




50Hz, Fundamental


150Hz, 3rd Harm.

Mag. = 0.5, Phase = 270°

500Hz, 10th Harm.

Mag. = 0.1, Phase = 0°

350Hz, 7th Harm.

Mag. = 0.4, Phase = 30°

5




50Hz, Fundamental


150Hz, 3rd Harm.

Mag. = 0.5, Phase = 270°

500Hz, 10th Harm.

Mag. = 0.1, Phase = 0°

350Hz, 7th Harm.

Mag. = 0.4, Phase = 30°

Reference signals

6




50Hz, Mag. = 3, Phase = 0°

150Hz, Mag. = 0.5, Phase = 270°

350Hz, Mag. = 0.4, Phase = 30°

500Hz, Mag. = 0.1, Phase = 0°

3

2

1

0

-1

-2

-3

6




50Hz, Mag. = 3, Phase = 0°

150Hz, Mag. = 0.5, Phase = 270°

350Hz, Mag. = 0.4, Phase = 30°

500Hz, Mag. = 0.1, Phase = 0°

3

2

1

0

-1

-2

-3

Raw signal

50Hz, Mag =1, Phase = 0°

Output??

Bandpass around 50Hz

6




3

2

1

0

-1

-2

-3

6




3

2

1

0

-1

-2

-3

3

2

1

0

-1

-2

-3

50Hz, Mag =3, Phase = 0°

Delay due to Filters

Harmonic estimation must take into account -

7




Robust against rogue harmonics

Buffer delay vs Accuracy

Robust against frequency error

Sample by sample update vs Buffered frame update

Sampling rate vs Harmonic range

Harmonic estimation must take into account -

7







Sample by sample update vs Buffered frame update


Calculations must finish within iteration of the real-time loop

Fourier based methods

• Discrete fourier transform (DFT)

Sliding Discrete Fourier Transform (SDFT)

Modulated Sliding Discrete Fourier Transform (mSDFT)

Time-frequency based methods

• Second Order Generalized Integrators (SOGI)

8




• Maximum sampling rate is 10kHz

• On each iteration, all of the calculations in

the model should be completed to prevent

data loss.

• Otherwise, the controller either skips the

model (overrun), or cannot react real time

(latency).

9

Real time processing on HIL targets

Frame based processing vs sample by sample

processing. input, fundamental component,

sample by sample (SDFT), frame based (DFT).



• Maximum sampling rate is 10kHz

• On each iteration, all of the calculations in

the model should be completed to prevent

data loss.

• Otherwise, the controller either skips the

model (overrun), or cannot react real time

(latency).

9


Frame based processing vs sample by sample

processing. input, fundamental component,

sample by sample (SDFT), frame based (DFT).



10




Input signal to test the real time harmonic estimation

Harmonic orders – 1st, 2nd, 3rd, 5th, 7th,11th, 13th

• Outputs are estimated waveforms; not just magnitudes with phase angles.

10


RT computation time results for SDFT, mSDFT and SOGI



100µs (10kHz) limit

for 1 iteration in RT


10


RT test results step change at 0.5s for SDFT, mSDFT and SOGI



50Hz

100Hz

150Hz 550Hz

350Hz

250Hz


• mSDFT outperformed the other methods in less burden and equally accurate results.

10


RT test results step change at 0.5s for SDFT, mSDFT and SOGI



50Hz

100Hz

150Hz 550Hz

350Hz

250Hz

• A compromise is made,

Fundamental components at 10kHz

Harmonic components at 2kHz

• It is possible to process even 146

measurements on a single PXI and obtain

phasors of the entire grid instantaneously.

• While still leaving considerable processing

power for other tasks.

11

CoSES PQ meter using mSDFT



• Various signal processing algorithms and data processing approaches are compared for

monitoring of CoSES

• An efficient, fast and robust PQ meter application using mSDFT has been implemented

• PQ meter has possibility to reduce controller burden even more by switching to buffered

frame updates if necessary

• Future work

Reducing frequency fluctuation dependency.

Using the PQ meter for feedback in a live experiment.

12

Conclusions and future work



Supplemenatary Slides

17





• Sliding Discrete Fourier Transform (SDFT)

• Modulated Sliding Discrete Fourier Transform (mSDFT)

7

Implementation nuances



Window length = 2 cycles

Signal arrives here Result arrives here

Short delay





7



Window length = 4 cycles (more information)

Signal arrives here Better quality result arrives here


Longer delay





7



Window length = 4 cycles (more information)

Signal arrives here Better quality result arrives here


Longer delay






8








8




Sampling rate = 10 kHz

Computation rate High






8






Computation rate medium





8






Computation rate low





8












8











9




One data frame





9




One data frame





9








9








9








9








9








10








10








10








10





Grid frequency estimation

must be accurate.


• Second Order generalized integrators

10




Measured Signal

Pre-determined group

of possible harmonics

Parallel SOGIs

Grid Frequency,

SOGI Gain tuning

Estimated harmonics


• Second Order generalized integrators

10





Measured Signal

Pre-determined group

of possible harmonics

Parallel SOGIs

Grid Frequency,

SOGI Gain tuning

Estimated harmonics

Documents

Fast harmonic estimation using real-time embedded